[mlir][GPUToSPIRV] Add a test pass to set workgroup size for kernel functions.

[mravishankar]

Summary

This allows using command line flags to lowere from GPU to SPIR-V. The
pass added is only for testing/example purposes. Most uses cases will
need more fine-grained control on setting workgroup sizes for kernel
functions.

Depends On D84590

[rriddle]

mlir/lib/Conversion/GPUToSPIRV/ConvertGPUToSPIRVPass.cpp

42

Please do not add test passes to the main tree, they should only be in the test/ directory.

[mravishankar]

Addressing comments

[mravishankar]

mlir/lib/Conversion/GPUToSPIRV/ConvertGPUToSPIRVPass.cpp

42

Moved the pass to test directory

[rriddle]

(Leaving approval to Lei or Stephan)

mlir/test/lib/Dialect/SPIRV/TestEntryPointAbi.cpp

28 ↗

(On Diff #281014)

This doesn't look used?

33 ↗

(On Diff #281014)

Why is this option int64_t if you only use it as int32_t?

[mravishankar]

Addressing comments

[mravishankar]

mlir/test/lib/Dialect/SPIRV/TestEntryPointAbi.cpp

28 ↗

(On Diff #281014)

Right. Removed it.

33 ↗

(On Diff #281014)

Changed to int32_t.

[herhut]

mlir/test/lib/Dialect/SPIRV/TestEntryPointAbi.cpp

28 ↗

(On Diff #281055)

Left-over comment?

[mravishankar]

Remove commented out code.

[mravishankar]

Adding missing new line at end of file.

mlir/lib/Conversion/GPUToSPIRV/ConvertGPUToSPIRVPass.cpp

 #include "mlir/Conversion/GPUToSPIRV/ConvertGPUToSPIRV.h"
 #include "mlir/Conversion/SCFToSPIRV/SCFToSPIRV.h"
 #include "mlir/Conversion/StandardToSPIRV/ConvertStandardToSPIRV.h"
 #include "mlir/Dialect/GPU/GPUDialect.h"
 #include "mlir/Dialect/SCF/SCF.h"
 #include "mlir/Dialect/SPIRV/SPIRVDialect.h"
 #include "mlir/Dialect/SPIRV/SPIRVLowering.h"
 #include "mlir/Dialect/SPIRV/SPIRVOps.h"
+#include "mlir/Pass/Pass.h"
 
 using namespace mlir;
 
 namespace {
 /// Pass to lower GPU Dialect to SPIR-V. The pass only converts the gpu.func ops
 /// inside gpu.module ops. i.e., the function that are referenced in
 /// gpu.launch_func ops. For each such function
 ///
 /// 1) Create a spirv::ModuleOp, and clone the function into spirv::ModuleOp
 /// (the original function is still needed by the gpu::LaunchKernelOp, so cannot
 /// replace it).
 ///
 /// 2) Lower the body of the spirv::ModuleOp.
 struct GPUToSPIRVPass : public ConvertGPUToSPIRVBase<GPUToSPIRVPass> {
   void runOnOperation() override;
 };
+
+/// Pass to set the spv.entry_point_abi
+class TestSpirvEntryPointABIPass
+    : public PassWrapper<TestSpirvEntryPointABIPass,
+                         OperationPass<gpu::GPUModuleOp>> {
+public:
+  TestSpirvEntryPointABIPass() = default;
+  TestSpirvEntryPointABIPass(ArrayRef<int64_t> sizes) { workgroupSize = sizes; }
+  TestSpirvEntryPointABIPass(const TestSpirvEntryPointABIPass &) {}
+  void runOnOperation() override;
+
+private:
+  Pass::ListOption<int64_t> workgroupSize{
+      *this, "workgroup-size",
+      llvm::cl::desc(
+          "Workgroup size to use for all gpu.func kernels in the module, "
+          "specified with x-dimension first, y-dimension next and z-dimension "
+          "last. Unspecified dimensions will be set to 1"),
+      llvm::cl::ZeroOrMore, llvm::cl::MiscFlags::CommaSeparated};
+};
 } // namespace
 
 void GPUToSPIRVPass::runOnOperation() {
   MLIRContext *context = &getContext();
   ModuleOp module = getOperation();
 
   SmallVector<Operation *, 1> kernelModules;
   OpBuilder builder(context);
void GPUToSPIRVPass::runOnOperation() {
   populateGPUToSPIRVPatterns(context, typeConverter, patterns);
   populateSCFToSPIRVPatterns(context, typeConverter,scfContext, patterns);
   populateStandardToSPIRVPatterns(context, typeConverter, patterns);
 
   if (failed(applyFullConversion(kernelModules, *target, patterns)))
     return signalPassFailure();
 }
 
+void TestSpirvEntryPointABIPass::runOnOperation() {
+  gpu::GPUModuleOp gpuModule = getOperation();
+  MLIRContext *context = &getContext();
+  StringRef attrName = spirv::getEntryPointABIAttrName();
+  for (gpu::GPUFuncOp gpuFunc : gpuModule.getOps<gpu::GPUFuncOp>()) {
+    if (!gpu::GPUDialect::isKernel(gpuFunc) || gpuFunc.getAttr(attrName))
+      continue;
+    SmallVector<int32_t, 3> workgroupSize32 = llvm::to_vector<3>(
+        llvm::map_range(workgroupSize, [](int64_t v) -> int32_t {
+          return static_cast<int32_t>(v);
+        }));
+    workgroupSize32.resize(3, 1);
+    gpuFunc.setAttr(attrName,
+                    spirv::getEntryPointABIAttr(workgroupSize32, context));
+  }
+}
+
 std::unique_ptr<OperationPass<ModuleOp>> mlir::createConvertGPUToSPIRVPass() {
   return std::make_unique<GPUToSPIRVPass>();
 }
+
+namespace mlir {
+void registerTestSpirvEntryPointABIPass() {
+  PassRegistration<TestSpirvEntryPointABIPass> registration(
+      "test-spirv-entry-point-abi",
+      "Set the spv.entry_point_abi attribute on GPU kernel function within the "
+      "module, intended for testing only");
+}
+} // namespace mlir

mlir/test/Conversion/GPUToSPIRV/test_spirv_entry_point.mlir

+// RUN: mlir-opt -test-spirv-entry-point-abi %s | FileCheck %s -check-prefix=DEFAULT
+// RUN: mlir-opt -test-spirv-entry-point-abi="workgroup-size=32" %s | FileCheck %s -check-prefix=WG32
+
+//      DEFAULT: gpu.func @foo()
+// DEFAULT-SAME: spv.entry_point_abi = {local_size = dense<1> : vector<3xi32>}
+
+//      WG32: gpu.func @foo()
+// WG32-SAME:  spv.entry_point_abi = {local_size = dense<[32, 1, 1]> : vector<3xi32>}
+
+gpu.module @kernels {
+  gpu.func @foo() kernel {
+    gpu.return
+  }
+}
 No newline at end of file

mlir/tools/mlir-opt/mlir-opt.cpp

 void registerTestLoopUnrollingPass();
 void registerTestMatchers();
 void registerTestMemRefDependenceCheck();
 void registerTestMemRefStrideCalculation();
 void registerTestOpaqueLoc();
 void registerTestPreparationPassWithAllowedMemrefResults();
 void registerTestReducer();
 void registerTestGpuParallelLoopMappingPass();
+void registerTestSpirvEntryPointABIPass();
 void registerTestSCFUtilsPass();
 void registerTestVectorConversions();
 void registerVectorizerTestPass();
 } // namespace mlir
 
 static cl::opt<std::string>
     inputFilename(cl::Positional, cl::desc("<input file>"), cl::init("-"));
 
#endif
   registerTestLoopUnrollingPass();
   registerTestMatchers();
   registerTestMemRefDependenceCheck();
   registerTestMemRefStrideCalculation();
   registerTestOpaqueLoc();
   registerTestPreparationPassWithAllowedMemrefResults();
   registerTestReducer();
   registerTestGpuParallelLoopMappingPass();
+  registerTestSpirvEntryPointABIPass();
   registerTestSCFUtilsPass();
   registerTestVectorConversions();
   registerVectorizerTestPass();
 }
 #endif
 
 static cl::opt<bool>
     showDialects("show-dialects",